STEPHEN F. SMITH
Research Professor, Robotics
www

My research interests are in artificial intelligence, primarily in the areas of constraint-based search and optimization, automated planning and scheduling, configurable and adaptive problem solving systems, multi-agent and multi-robot coordination, mixed-initiative decision-making, and naturally inspired search procedures. One integrating focus has been the development of core technologies for coordination and control of large-scale, multi-actor systems, and their application in domains such as manufacturing, transportation, logistics, space mission planning, and eBusiness.

Practical Planning and Scheduling – I am interested broadly in the theory and practice of next-generation technologies for practical planning and scheduling. One thread of current research focuses on algorithms for constructing robust plans and schedules, which capture sets of possible execution futures and anticipate executional uncertainty. Another topic of current research is over-subscribed planning and scheduling, which involves problems where available resources preclude accomplishment off all goals and solutions must therefore maximize expected gain. Other general interests here include constraint-based planning and scheduling, integrated action selection and resource allocation, visualization and mixed-initiative manipulation of plans/schedules, reactive plan/schedule repair, planning and scheduling under complex (and potentially conflicting) constraints, and planning/scheduling search-space analysis.

Coordinating Distributed Planning and Scheduling Agents – Increasingly, my research has emphasized planning and scheduling problems that are inherently distributed and require mechanisms for coordinated decision-making by multiple agents. One current focus (within the Coordinators project) is on distributed execution and management of joint schedules in an uncertain execution environment, where each agent has responsibility for carrying out some portion of an overall operation, the actions of different agents are inter-dependent, but no one agent has a complete global view. Other current work is investigating algorithms for distributed use of 3D space over time by multiple air vehicles. Broader research interests here include self-scheduling systems, distributed constraint optimization, and negotiation-based approaches to distributed planning and scheduling.

Adaptive and Configurable Problem Solvers – Another generalresearch interest is the design of configurable and adaptive systems. One area of current research focuses on mechanisms for exploiting the use of multiple heuristics to efficiently solve planning, scheduling and optimization problems. Topics here include online learning strategies for allocating trials to heuristics within iterative sampling search procedures, algorithm portfolio design, and other adaptive search procedures. At another level, I am also interested in the development of reconfigurable planning and scheduling system architectures, which promote rapid development of high performance application systems.

For more information and publications, please visit http://www.ozone.ri.cmu.edu/.